Best Practices Guide: AI Agents
Updated May 28, 2026
On this page
- What makes a good Agent prompt
- Simple prompt template
- Tips for better Agent prompts
- Before you build
- Admins
- What Admins should build
- What good Admin Agents look like
- Admin examples customers can adapt
- Best practices for Admins
- Managers
- What Managers should build
- What good Manager Agents look like
- Manager examples customers can adapt
- Best practices for Managers
- Users
- What Users should build
- What good User Agents look like
- User examples customers can adapt
- Best practices for Users
- Department examples by role
- Common mistakes to avoid
- What to know about triggers
- Rollout advice
Demodesk's AI Agents help your team automate recurring work across the sales process. They can prepare reps before meetings, analyze meetings after they've been processed in Demodesk, monitor risks, summarize patterns, and send updates without manual follow-up.
This guide helps you turn AI Agents into practical workflows for Admins, Managers, and individual Users.
What makes a good Agent prompt
A good Agent prompt should clearly define five things:
| # | Element | What to define | Example |
|---|---|---|---|
| 1 | Trigger | When the Agent should run | Every Monday morning, after a meeting has been processed in Demodesk, or when triggered via webhook |
| 2 | Scope | What the Agent should analyze | Meeting recordings, AI summaries, CRM context, past meetings, or connected tools |
| 3 | Task | What the Agent should do | Review, compare, rank, summarize, flag, prepare, or update |
| 4 | Output | What the result should include | A short summary, ranked risks, action items, snippets, or a structured brief |
| 5 | Action | What should happen after the result is generated | Post to Demodesk chat, email it to yourself, push to Slack, Microsoft Teams, Notion, Jira, GitLab, or a CRM through an API |
Simple prompt template
Trigger Run after [event / schedule / webhook].
Scope Focus on [meetings / accounts / reps / time range / meeting type].
Task Analyze [sources]. Look for [signals]. Compare against [criteria].
Output Return [format, length, fields, ranking logic].
Action Send the result to [Demodesk chat / email to yourself / connected system].
Tips for better Agent prompts
| Tip | Why it helps |
|---|---|
| Keep prompts short and specific | Five to ten lines is often enough |
| Tell the Agent exactly what "good" looks like | Helps produce concise, specific, account-grounded output |
| Use a clear format | Bullets, tables, top 3 risks, next steps, or structured briefs are easier to review |
| Start with one simple workflow | Keeps the Agent narrow enough to validate and improve |
| Specify which meetings matter | Useful when the Agent acts on discovery, demo, handover, renewal, QBR, escalation, or onboarding meetings |
| Review execution history | Helps tighten prompts that are too broad or generic |
Before you build
AI Agents work best when Demodesk has access to the right context.
| Strong starting points | Additional context that can make Agents more powerful |
|---|---|
| Demodesk meeting data | Connected APIs |
| AI summaries and transcripts | Slack, Microsoft Teams, Notion, Jira, GitLab, or CRM integrations |
| CRM context | Webhook-based triggers from external systems |
| Past meetings with the same customer |
Admins
Admins usually set up the foundation for the team. They define shared workflows, control access, connect systems, and create Agents that improve consistency across teams.
What Admins should build
| Admin workflow | Purpose |
|---|---|
| Team-wide post-meeting workflows | Improve consistency across teams |
| Shared digests for leadership or enablement | Keep key stakeholders informed |
| Pipeline-risk monitoring | Identify risks and gaps |
| Cross-team pattern analysis | Surface recurring themes |
| Webhook-based automations | Connect Agents to internal tools |
What good Admin Agents look like
| Quality | What it means |
|---|---|
| Clear trigger | The Agent runs at the right time |
| Defined audience | The output is written for the right people |
| Standardized output format | Results are easier to compare |
| Narrow scope | The Agent is reliable |
| Useful before advanced actions | The workflow works even before API actions are added |
Admin examples customers can adapt
1. Weekly Pipeline Risk & CRM Quality Brief
Trigger: Run every Monday at 8:00 AM.
Scope: Review processed external meetings from the last 7 days for our Sales team, with CRM context where available.
Task: Identify pipeline risks, missing CRM details, unclear next steps, stalled opportunities, and weak qualification signals.
Output: Return:
Top pipeline risks
Incomplete or inconsistent CRM fields
Opportunities with unclear next steps
Recommended follow-up actions for managers or RevOps
Action: Post to Demodesk chat and email the brief to me.
2. Weekly Leadership Account Brief
Trigger: Run every Monday at 7:00 AM.
Scope: Review key external meetings from the last 7 days across strategic accounts.
Task: Summarize the most important commercial themes, including risks, momentum shifts, customer priorities, and expansion signals.
Output: Return:
Top wins
Top risks
Strategic account developments
Recurring customer themes
Actions leadership should be aware of
Action: Post the brief to a Slack channel (e.g. #leadership-weekly) or a Microsoft Teams group via the connected integration.
3. Weekly Market Signal Digest
Trigger: Run every Friday at 4:00 PM.
Scope: Analyze external meetings from the past week.
Task: Identify recurring objections, product feedback themes, and competitor references.
Output: Return:
Top objections raised
Product feedback themes
Competitor references and context
Notable customer quotes
Summary of what changed from the previous week
Action: Append the digest as a new entry in a Notion database (one row per week) so the team can track trends over time.
4. External Request Routing Agent
Trigger: Run when triggered via webhook.
Scope: Use the webhook payload and any matching customer or meeting context.
Task: Validate the incoming information, classify the request, and prepare the required handoff summary or follow-up package.
Output: Return:
Request type
Relevant customer or project reference
Summary of key details
Missing information
Recommended next action
Action: Push to the connected system via API. For example: create a Jira ticket for engineering escalations, open a GitLab or GitHub issue for product feedback, or send a structured alert to a Slack or Microsoft Teams channel.
Best practices for Admins
| Best practice | Guidance |
|---|---|
| Start with shared workflows | Prioritize Agents that benefit multiple users |
| Standardize output formats | Make results easier for managers to compare |
| Pilot with one team | Validate before enabling company-wide |
| Begin with read-and-notify workflows | Add API actions only after the logic works |
| Monitor cost, usage, and reliability | Keep rollout controlled |
Managers
Managers should use AI Agents to scale coaching, improve deal inspection, and keep visibility high without reading every meeting manually.
What Managers should build
| Manager workflow | Purpose |
|---|---|
| Rep coaching summaries | Identify coaching opportunities |
| Deal risk monitoring | Surface deals that need attention |
| Pipeline inspection digests | Improve visibility across the team |
| Follow-up quality checks | Check whether next steps are clear |
| Onboarding or handover quality reviews | Improve transition quality |
What good Manager Agents look like
| Quality | What it means |
|---|---|
| One coaching or inspection goal | The output stays focused |
| Easy to scan team-wide patterns | Managers can quickly identify trends |
| Attention-oriented output | The Agent highlights what needs action |
| Faster intervention | Managers can follow up sooner |
Manager examples customers can adapt
1. Demo Review Assistant
Trigger: Run after a meeting has been processed in Demodesk.
Scope: Only review demo meetings from my team.
Task: Evaluate how well the rep handled customer context, requirements, objections, and next steps.
Output: Return:
What went well
What was missing
One coaching recommendation
Whether follow-up was clearly defined
Action: Post to Demodesk chat so the rep sees the feedback in context.
2. Weekly Coaching Trends Summary
Trigger: Run every Friday at 5:00 PM.
Scope: Review my team's processed meetings from this week.
Task: Analyze recurring strengths, weaknesses, and coaching opportunities across recent conversations.
Output: Return:
Most common improvement areas
Strongest execution patterns
Reps who may need support
Coaching priorities for next week
Action: Email the summary to me, and post it to a private Slack or Microsoft Teams channel for my managers' circle.
3. Customer Requirements Tracker
Trigger: Run after a meeting has been processed in Demodesk.
Scope: Only review discovery and demo meetings for my team.
Task: Extract customer requirements, buying criteria, blockers, timeline signals, and decision drivers.
Output: Return:
Account
Main requirements
Blockers
Urgency level
Missing information
Recommended follow-up questions
Action: Push the structured output to the CRM opportunity record via API (e.g. HubSpot or Salesforce) so deal context stays in one place.
4. Handover Quality Review
Trigger: Run after a meeting has been processed in Demodesk.
Scope: Only review Customer Success handover or onboarding meetings.
Task: Check whether customer goals, stakeholders, success metrics, blockers, and next steps were clearly captured.
Output: Return:
Completeness score
Missing items
Suggested follow-up questions
Summary for the manager
Action: Post to a dedicated Slack handover channel and email me a copy.
Best practices for Managers
| Best practice | Guidance |
|---|---|
| Don't ask for everything | Ask for one decision-ready output |
| Use meeting-type scope | Keep the Agent focused |
| Ask for rankings and evidence | Avoid generic summaries |
| Keep output short | Make it reviewable in under 2 minutes |
| Combine recurring and event-based Agents | Use digests for patterns and post-meeting Agents for immediate follow-up |
Users
Individual users should focus on Agents that help them prepare better, follow up faster, and stay on top of their own accounts.
What Users should build
| User workflow | Purpose |
|---|---|
| Pre-meeting prep briefs | Prepare faster and better |
| Account monitoring | Stay on top of account changes |
| Personal coaching snapshots | Improve individual performance |
| Stakeholder tracking | Keep customer context clear |
| Follow-up and action-item reminders | Reduce missed next steps |
What good User Agents look like
| Quality | What it means |
|---|---|
| Saves time | Helps before or after meetings |
| Improves consistency | Makes preparation and follow-up repeatable |
| Personal and narrow | Focuses on the user's own meetings or accounts |
| Fits into a daily workflow | Supports regular work without extra effort |
User examples customers can adapt
1. Pre-Meeting Account Brief
Trigger: Run 1 hour before my scheduled external meetings.
Scope: Use the upcoming meeting, CRM context, and past meetings with the same customer.
Task: Prepare a concise meeting brief with account context, open risks, previous commitments, likely objections, and recommended focus areas.
Output: Return:
Account summary
Stakeholders
Last meeting takeaways
Open next steps
3 questions I should ask
2 likely objections to prepare for
Action: Email the brief to me 1 hour before the meeting.
2. Follow-Up Action Extractor
Trigger: Run after a meeting has been processed in Demodesk.
Scope: Only my external meetings.
Task: Extract promised next steps, owners, dates, unanswered questions, and follow-up risks.
Output: Return:
My action items
Customer action items
Due dates
Unresolved questions
A short suggested follow-up summary
Action: Post action items to Demodesk chat, and prepare a follow-up email as a draft in my inbox (coming soon) that I can review and send manually.
3. Weekly Renewal Check-In
Trigger: Run every Friday at 3:00 PM.
Scope: Review my renewal-related customer meetings from the last 7 days.
Task: Summarize account sentiment, open risks, unresolved requests, renewal signals, and next steps.
Output: Return:
Customer health summary
Renewal risks
Open requests
Actions I should take next
Accounts that need urgent attention
Action: Append the check-in to my personal Notion renewal tracker so I can review trends across weeks.
4. Friday Customer Highlights Digest
Trigger: Run every Friday at 4:30 PM.
Scope: Review my processed customer meetings from this week.
Task: Find positive feedback, successful outcomes, praise for product or service, and strong customer quotes.
Output: Return:
Top positive customer moments
Customer quotes
Accounts with strong momentum
Themes worth sharing internally
Action: Post the digest to a Slack or Microsoft Teams channel (e.g. #customer-love) so the team can celebrate wins together.
Best practices for Users
| Best practice | Guidance |
|---|---|
| Build one prep Agent and one follow-up Agent first | Start with the highest-frequency workflows |
| Use a narrow audience | Focus on your meetings, your accounts, or one meeting type |
| Prefer actionable outputs | Avoid long summaries |
| Ask what changed since the last meeting | Make the output more useful |
| Refine after 2 to 3 runs | Improve based on real execution history |
Department examples by role
| Department | Admin examples | Manager examples | User examples |
|---|---|---|---|
| Sales | Pipeline risk brief, market signal digest | Demo review assistant, customer requirements tracker | Pre-meeting account brief, follow-up action extractor |
| Customer Success | Leadership account brief, external request routing | Handover quality review, weekly coaching trends summary | Weekly renewal check-in, Friday customer highlights digest |
| RevOps / Enablement | Pipeline risk and CRM quality monitoring | Coaching trends and process review | Prep briefs and action tracking |
| Product / Leadership | Market signal digest, leadership account brief | Account theme summaries and escalations review | Customer highlights and product feedback collection |
Common mistakes to avoid
| Mistake | Better approach |
|---|---|
| Asking the Agent to do too many things in one prompt | Keep each Agent focused on one outcome |
| Not defining which meetings or accounts matter | Specify the relevant meeting type, account set, team, or time range |
| Asking for a long narrative | Request a useful format such as bullets, tables, risks, or next steps |
| Starting with API actions before validating the logic | Validate the result first in Demodesk chat |
| Using vague goals like "analyze my meetings" or "find insights" | Define the exact signals and output format |
| Forgetting to specify the notification channel | State whether the result should go to chat, email, Slack, Microsoft Teams, Notion, or a connected tool |
What to know about triggers
| Trigger type | Best use case |
|---|---|
| Schedule | Digests and pattern analysis |
| After a meeting has been processed in Demodesk | Coaching, follow-up, and inspection |
| Webhook from an external system | External workflow orchestration |
Rollout advice
| Step | Action |
|---|---|
| 1 | Start with one Admin Agent, one Manager Agent, and one User Agent |
| 2 | Keep each one focused on a single outcome |
| 3 | Validate the output in Demodesk chat first |
| 4 | Tighten the prompt if results are too broad or repetitive |
| 5 | Add email, Slack, Microsoft Teams, Notion, or API actions only after the logic works reliably |
| 6 | Review execution history regularly and refine the prompt over time |
Final thought The best AI Agents are not the most complex ones. They are the ones that run at the right time, use the right context, and produce a clear output that helps someone act faster.
Related articles
More in this category
Still need a hand?
Our support team answers in the app chat, or at support@demodesk.com.